From: Corey Minyard <cminyard@xxxxxxxxxx>
ring_buffer_resize() would call rb_update_pages() with preempt
disabled. It did this with preempt disabled, bu rb_update_pages
would eventually claim the zone semaphore, which is a mutex in PREEMPT_RT,
thus resulting in a "scheduling while atomic" BUG.
When doing this, all other CPUs except the running one did the operation
in a work queue. So just do the current CPU operation in a work queue,
too. This simplifies the code, gets rid of the need for preempt_disable(),
and solves the RT issue.
Signed-off-by: Corey Minyard <cminyard@xxxxxxxxxx>
---
kernel/trace/ring_buffer.c | 10 +---------
1 file changed, 1 insertion(+), 9 deletions(-)
This isn't well tested, but fixes an obvious PREEMPT_RT problem and seems
like a good simplification. You can see the PREEMPT_RT bug if you enable
ftrace and all the self tests; it will fail running selftests when it tries
to resize the buffers.
diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index fc4da2d..972125e 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -1688,21 +1688,13 @@ int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size,
continue;
/* The update must run on the CPU that is being updated. */
- preempt_disable();
- if (cpu == smp_processor_id() || !cpu_online(cpu)) {
+ if (!cpu_online(cpu)) {
rb_update_pages(cpu_buffer);
cpu_buffer->nr_pages_to_update = 0;
} else {
- /*
- * Can not disable preemption for schedule_work_on()
- * on PREEMPT_RT.
- */
- preempt_enable();
schedule_work_on(cpu,
&cpu_buffer->update_pages_work);
- preempt_disable();
}
- preempt_enable();
}
/* wait for all the updates to complete */
--
1.8.3.1
--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
